Brreg

Brreg — Norwegian company register MCP server for Claude, Cursor and any MCP client

Brreg gives an AI assistant Brønnøysundregistrene's Enhetsregisteret as structured data: look an organisation up by organisasjonsnummer, turn a name into candidates without a silent pick, search with filters, regular expressions or a point and a radius, find who is registered near a place and how far apart two are, and see how organisations belong together — with key financials from Regnskapsregisteret, and an industry benchmark from Statistics Norway, when you ask for them. It runs as a hosted MCP server over HTTPS.

Connect

{ "mcpServers": { "brreg": { "type": "http", "url": "https://brreg.publifye.com/mcp" } } }

Claude Code: claude mcp add --transport http brreg https://brreg.publifye.com/mcp, then /mcp to sign in.

The OAuth flow runs in the browser on first use, with a Publifye account. Access is granted through that sign-in only; there is no API key to create, copy or rotate. The discovery chain is the same as for the other servers: [../../docs/connect.md](../../docs/connect.md).

The register

The edition the live service was serving on 2026-09-29, acquired that morning from Brønnøysundregistrene's bulk files:

The register is rebuilt from the full files every week. An entry Brønnøysundregistrene removes from open data is dropped within 24 hours. snapshot_status returns the age and counts of whichever edition is current, and every successful result names the edition it came from.

A number is exact; a name is not

entity_lookup takes an organisasjonsnummer, checks its checksum, and is never fuzzy.

entity_resolve takes a name and returns candidates, with resolution set to unique, ambiguous or unresolved and the order of the candidates stated. It never claims the first candidate is the organisation. Norwegian spellings are compared through one documented key — æ, ø and å folded, legal forms such as AS and ASA stripped, former names matched — so the comparison can be checked rather than trusted.

No language model is involved in answering a call. Every successful result carries the licence attribution and names the edition it came from, so an answer can be traced back to its source.

What the tools do

Every tool is documented with its exact description, annotations and input schema — 9 in all, generated from the service's own tools/list, never written by hand.

Machine-readable: [tools.json](tools.json) carries all 9 callable tools (9 capability, 0 session/cache) with full JSON Schema, plus every excluded bucket listed by name so the count is auditable.

All nine are read-only. Two are marked open-world: entity_lookup, which with live=true checks the entry against data.brreg.no, and entity_financials, which goes to Regnskapsregisteret for figures the service does not yet hold.

Four ways to narrow a search

entity_search takes at least one of query, address, website, match or a filter, and everything you give is AND'ed together.

Filters. kind, org_form, municipality, postcode, city, nace, sector, status, public_body, vat_registered, parent_orgnr, employees_min/max, registered_from/to. The values come from code_list, so a filter is a code you looked up, not a guess.

Regular expressions. match takes 1 to 4 {field, pattern} terms over activity, address, city, email, name, phone or website. A pattern is RE2, at most 200 bytes, matched case-insensitively against the normalised field value — NFC, lower-cased, whitespace-collapsed and truncated to 256 bytes, so a pattern anchored deep into a long activity text may find nothing — and you anchor it yourself: ^post matches at the start, an unanchored pattern matches anywhere, and . tests that the field is present at all. One pattern per field; an invalid, over-long, over-complex or empty-string-matching pattern rejects the whole call with invalid_filter before any work is done. snapshot_status lists the fields the installed edition can match, in match_fields.

// generic company inboxes in a given industry

{ "nace": "94.91", "match": [ { "field": "email", "pattern": "^(post|info|kontakt|firmapost)[.0-9]*@" } ] }

A point and a radius. lat, lon and radius_km — all three or none — list the organisations whose registered address falls inside the circle, nearest first, each with distance_km and the address point it was measured from. Give a name query as well and the order stays relevance, with the radius acting as a filter. The radius is at most 10 km on its own, or 50 km together with another filter, and the centre must fall within latitude 57.5 to 81.5 and longitude 4 to 32.

The point is the Kartverket address point of the registered address — the business address for a main unit, the location address for a sub-unit. It is not a building footprint and not necessarily where the work is done. Sharing a point is the norm rather than the exception — an accountant's office, a shared business park — so each result says how many records share its point. Addresses given as c/o, post boxes, foreign addresses and addresses that do not match the address register get no point and never appear in a radius search.

Cursor. Paging has no depth limit: follow the cursor to the end of the result set. offset is a first-call convenience only, capped at 10,000.

Near a place, and how far

entity_nearby lists the organisations registered within a radius of a point, nearest first. The centre is either lat and lon or an orgnr — that organisation's own registered address, so "who is near this company" is one call; the anchor itself is left out of its results. The same caps apply as for a radius search: 10 km on its own, 50 km with another filter.

entity_distance measures from one place to up to 50 others and answers nearest first. A place is an orgnr or a lat/lon pair. The number is the straight-line distance between two registered address points — not a driving or walking distance and not a travel time; across a fjord the road can be many times longer. A place with no usable point is still answered, with located: false and the reason, never silently dropped.

Money: entity_financials

entity_financials is the tool for figures. include chooses key_figures (the default), filings — the years actually filed, which also answers for banks and insurers, whose key figures Regnskapsregisteret does not serve — and sector_benchmark, which sets the figures against Statistics Norway's statistics for the industry and size band. The benchmark is per industry, never per company, and names the level that answered. Figures are fetched from Regnskapsregisteret only when this tool asks, then stored; a year already held is not fetched again. document_year hands over that year's filed accounts as an expiring link (see below).

What a lookup can add

entity_lookup returns the register entry. Ask for more in fields — financials and filings are still accepted here, fetched only when asked for, though entity_financials is the front door for them:

document_year returns that year's filed annual accounts as an expiring private link — a scanned image with no text layer, not structured figures, and not a second route to them. The filing itself never travels in an MCP response: what comes back is a link, a size, a checksum and an expiry.

unit and buildings describe the parcel and the buildings at the registered address. They are not a statement about property the organisation owns or occupies, and there is no owner data anywhere in the service.

What it does not do

• No roles or persons — no board members, managing directors or owners.
• Not the authoritative register. Brreg is operated by Publifye AS and is not affiliated with Brønnøysundregistrene. For anything legally binding, use the official record at https://virksomhet.brreg.no.
• Not a reidentification tool. It is a register of organisations and cannot be used to look people up; there are no address-only lookups of individuals.
• No bulk export. A search answers a page at a time, followed by cursor; there is no download of the register.
• Sole proprietorships are held back. E-mail and phone numbers, former names and activity text are not shown; they get no map point, no parcel and no buildings; and a search without a name never lists them. A match term cannot reach one either — personal-data records contribute no values to the fields match reads.
• No structured annual accounts. Key figures and the list of filed years are structured data; the accounts themselves are not. A filed year can be handed over as a scanned copy on request, and a scan is where it ends — there are no machine-readable notes, line items or auditor's reports.
• No property ownership. Nothing from grunnboken; the parcel is the one at the registered address.

Data and licence

[PROVENANCE.md](PROVENANCE.md) records each dataset, how it reaches the service, how fresh it is, and what is withheld. Two licences apply, and a result carries the ones its own content needs.

Enhetsregisteret and Regnskapsregisteret are open data from Brønnøysundregistrene under the Norwegian Licence for Open Government Data (NLOD 2.0). Every successful result carries the attribution the licence asks for:

Inneholder data under Norsk lisens for offentlige data (NLOD) tilgjengeliggjort av Brønnøysundregistrene.

Address points, parcels and buildings come from Kartverket's Matrikkelen under Creative Commons Attribution 4.0 (CC BY 4.0). A result that carries a coordinate or a distance carries the Kartverket attribution too, and says what Publifye changed — the points were matched to each organisation's registered address and do not show where a business physically operates. A result that carries a parcel or buildings also carries the private-register notice that utleveringsforskriften § 5 tenth paragraph